fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/fs_struct.h>
  85 #include <linux/slab.h>
  86 #ifdef CONFIG_HARDWALL
  87 #include <asm/hardwall.h>
  88 #endif
  89 #include "internal.h"
  90
  91 /* NOTE:
  92  *      Implementing inode permission operations in /proc is almost
  93  *      certainly an error.  Permission checks need to happen during
  94  *      each system call not at open time.  The reason is that most of
  95  *      what we wish to check for permissions in /proc varies at runtime.
  96  *
  97  *      The classic example of a problem is opening file descriptors
  98  *      in /proc for a task before it execs a suid executable.
  99  */
 100
 101 struct pid_entry {
 102         char *name;
 103         int len;
 104         mode_t mode;
 105         const struct inode_operations *iop;
 106         const struct file_operations *fop;
 107         union proc_op op;
 108 };
 109
 110 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 111         .name = (NAME),                                 \
 112         .len  = sizeof(NAME) - 1,                       \
 113         .mode = MODE,                                   \
 114         .iop  = IOP,                                    \
 115         .fop  = FOP,                                    \
 116         .op   = OP,                                     \
 117 }
 118
 119 #define DIR(NAME, MODE, iops, fops)     \
 120         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 121 #define LNK(NAME, get_link)                                     \
 122         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 123                 &proc_pid_link_inode_operations, NULL,          \
 124                 { .proc_get_link = get_link } )
 125 #define REG(NAME, MODE, fops)                           \
 126         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 127 #define INF(NAME, MODE, read)                           \
 128         NOD(NAME, (S_IFREG|(MODE)),                     \
 129                 NULL, &proc_info_file_operations,       \
 130                 { .proc_read = read } )
 131 #define ONE(NAME, MODE, show)                           \
 132         NOD(NAME, (S_IFREG|(MODE)),                     \
 133                 NULL, &proc_single_file_operations,     \
 134                 { .proc_show = show } )
 135
 136 /*
 137  * Count the number of hardlinks for the pid_entry table, excluding the .
 138  * and .. links.
 139  */
 140 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 141         unsigned int n)
 142 {
 143         unsigned int i;
 144         unsigned int count;
 145
 146         count = 0;
 147         for (i = 0; i < n; ++i) {
 148                 if (S_ISDIR(entries[i].mode))
 149                         ++count;
 150         }
 151
 152         return count;
 153 }
 154
 155 static int get_task_root(struct task_struct *task, struct path *root)
 156 {
 157         int result = -ENOENT;
 158
 159         task_lock(task);
 160         if (task->fs) {
 161                 get_fs_root(task->fs, root);
 162                 result = 0;
 163         }
 164         task_unlock(task);
 165         return result;
 166 }
 167
 168 static int proc_cwd_link(struct inode *inode, struct path *path)
 169 {
 170         struct task_struct *task = get_proc_task(inode);
 171         int result = -ENOENT;
 172
 173         if (task) {
 174                 task_lock(task);
 175                 if (task->fs) {
 176                         get_fs_pwd(task->fs, path);
 177                         result = 0;
 178                 }
 179                 task_unlock(task);
 180                 put_task_struct(task);
 181         }
 182         return result;
 183 }
 184
 185 static int proc_root_link(struct inode *inode, struct path *path)
 186 {
 187         struct task_struct *task = get_proc_task(inode);
 188         int result = -ENOENT;
 189
 190         if (task) {
 191                 result = get_task_root(task, path);
 192                 put_task_struct(task);
 193         }
 194         return result;
 195 }
 196
 197 static struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 198 {
 199         struct mm_struct *mm;
 200         int err;
 201
 202         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 203         if (err)
 204                 return ERR_PTR(err);
 205
 206         mm = get_task_mm(task);
 207         if (mm && mm != current->mm &&
 208                         !ptrace_may_access(task, mode)) {
 209                 mmput(mm);
 210                 mm = ERR_PTR(-EACCES);
 211         }
 212         mutex_unlock(&task->signal->cred_guard_mutex);
 213
 214         return mm;
 215 }
 216
 217 struct mm_struct *mm_for_maps(struct task_struct *task)
 218 {
 219         return mm_access(task, PTRACE_MODE_READ);
 220 }
 221
 222 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 223 {
 224         int res = 0;
 225         unsigned int len;
 226         struct mm_struct *mm = get_task_mm(task);
 227         if (!mm)
 228                 goto out;
 229         if (!mm->arg_end)
 230                 goto out_mm;    /* Shh! No looking before we're done */
 231
 232         len = mm->arg_end - mm->arg_start;
 233
 234         if (len > PAGE_SIZE)
 235                 len = PAGE_SIZE;
 236
 237         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 238
 239         // If the nul at the end of args has been overwritten, then
 240         // assume application is using setproctitle(3).
 241         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 242                 len = strnlen(buffer, res);
 243                 if (len < res) {
 244                     res = len;
 245                 } else {
 246                         len = mm->env_end - mm->env_start;
 247                         if (len > PAGE_SIZE - res)
 248                                 len = PAGE_SIZE - res;
 249                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 250                         res = strnlen(buffer, res);
 251                 }
 252         }
 253 out_mm:
 254         mmput(mm);
 255 out:
 256         return res;
 257 }
 258
 259 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 260 {
 261         struct mm_struct *mm = mm_for_maps(task);
 262         int res = PTR_ERR(mm);
 263         if (mm && !IS_ERR(mm)) {
 264                 unsigned int nwords = 0;
 265                 do {
 266                         nwords += 2;
 267                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 268                 res = nwords * sizeof(mm->saved_auxv[0]);
 269                 if (res > PAGE_SIZE)
 270                         res = PAGE_SIZE;
 271                 memcpy(buffer, mm->saved_auxv, res);
 272                 mmput(mm);
 273         }
 274         return res;
 275 }
 276
 277
 278 #ifdef CONFIG_KALLSYMS
 279 /*
 280  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 281  * Returns the resolved symbol.  If that fails, simply return the address.
 282  */
 283 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 284 {
 285         unsigned long wchan;
 286         char symname[KSYM_NAME_LEN];
 287
 288         wchan = get_wchan(task);
 289
 290         if (lookup_symbol_name(wchan, symname) < 0)
 291                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 292                         return 0;
 293                 else
 294                         return sprintf(buffer, "%lu", wchan);
 295         else
 296                 return sprintf(buffer, "%s", symname);
 297 }
 298 #endif /* CONFIG_KALLSYMS */
 299
 300 static int lock_trace(struct task_struct *task)
 301 {
 302         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 303         if (err)
 304                 return err;
 305         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 306                 mutex_unlock(&task->signal->cred_guard_mutex);
 307                 return -EPERM;
 308         }
 309         return 0;
 310 }
 311
 312 static void unlock_trace(struct task_struct *task)
 313 {
 314         mutex_unlock(&task->signal->cred_guard_mutex);
 315 }
 316
 317 #ifdef CONFIG_STACKTRACE
 318
 319 #define MAX_STACK_TRACE_DEPTH   64
 320
 321 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 322                           struct pid *pid, struct task_struct *task)
 323 {
 324         struct stack_trace trace;
 325         unsigned long *entries;
 326         int err;
 327         int i;
 328
 329         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 330         if (!entries)
 331                 return -ENOMEM;
 332
 333         trace.nr_entries        = 0;
 334         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 335         trace.entries           = entries;
 336         trace.skip              = 0;
 337
 338         err = lock_trace(task);
 339         if (!err) {
 340                 save_stack_trace_tsk(task, &trace);
 341
 342                 for (i = 0; i < trace.nr_entries; i++) {
 343                         seq_printf(m, "[<%pK>] %pS\n",
 344                                    (void *)entries[i], (void *)entries[i]);
 345                 }
 346                 unlock_trace(task);
 347         }
 348         kfree(entries);
 349
 350         return err;
 351 }
 352 #endif
 353
 354 #ifdef CONFIG_SCHEDSTATS
 355 /*
 356  * Provides /proc/PID/schedstat
 357  */
 358 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 359 {
 360         return sprintf(buffer, "%llu %llu %lu\n",
 361                         (unsigned long long)task->se.sum_exec_runtime,
 362                         (unsigned long long)task->sched_info.run_delay,
 363                         task->sched_info.pcount);
 364 }
 365 #endif
 366
 367 #ifdef CONFIG_LATENCYTOP
 368 static int lstats_show_proc(struct seq_file *m, void *v)
 369 {
 370         int i;
 371         struct inode *inode = m->private;
 372         struct task_struct *task = get_proc_task(inode);
 373
 374         if (!task)
 375                 return -ESRCH;
 376         seq_puts(m, "Latency Top version : v0.1\n");
 377         for (i = 0; i < 32; i++) {
 378                 struct latency_record *lr = &task->latency_record[i];
 379                 if (lr->backtrace[0]) {
 380                         int q;
 381                         seq_printf(m, "%i %li %li",
 382                                    lr->count, lr->time, lr->max);
 383                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 384                                 unsigned long bt = lr->backtrace[q];
 385                                 if (!bt)
 386                                         break;
 387                                 if (bt == ULONG_MAX)
 388                                         break;
 389                                 seq_printf(m, " %ps", (void *)bt);
 390                         }
 391                         seq_putc(m, '\n');
 392                 }
 393
 394         }
 395         put_task_struct(task);
 396         return 0;
 397 }
 398
 399 static int lstats_open(struct inode *inode, struct file *file)
 400 {
 401         return single_open(file, lstats_show_proc, inode);
 402 }
 403
 404 static ssize_t lstats_write(struct file *file, const char __user *buf,
 405                             size_t count, loff_t *offs)
 406 {
 407         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 408
 409         if (!task)
 410                 return -ESRCH;
 411         clear_all_latency_tracing(task);
 412         put_task_struct(task);
 413
 414         return count;
 415 }
 416
 417 static const struct file_operations proc_lstats_operations = {
 418         .open           = lstats_open,
 419         .read           = seq_read,
 420         .write          = lstats_write,
 421         .llseek         = seq_lseek,
 422         .release        = single_release,
 423 };
 424
 425 #endif
 426
 427 static int proc_oom_score(struct task_struct *task, char *buffer)
 428 {
 429         unsigned long points = 0;
 430
 431         read_lock(&tasklist_lock);
 432         if (pid_alive(task))
 433                 points = oom_badness(task, NULL, NULL,
 434                                         totalram_pages + total_swap_pages);
 435         read_unlock(&tasklist_lock);
 436         return sprintf(buffer, "%lu\n", points);
 437 }
 438
 439 struct limit_names {
 440         char *name;
 441         char *unit;
 442 };
 443
 444 static const struct limit_names lnames[RLIM_NLIMITS] = {
 445         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 446         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 447         [RLIMIT_DATA] = {"Max data size", "bytes"},
 448         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 449         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 450         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 451         [RLIMIT_NPROC] = {"Max processes", "processes"},
 452         [RLIMIT_NOFILE] = {"Max open files", "files"},
 453         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 454         [RLIMIT_AS] = {"Max address space", "bytes"},
 455         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 456         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 457         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 458         [RLIMIT_NICE] = {"Max nice priority", NULL},
 459         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 460         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 461 };
 462
 463 /* Display limits for a process */
 464 static int proc_pid_limits(struct task_struct *task, char *buffer)
 465 {
 466         unsigned int i;
 467         int count = 0;
 468         unsigned long flags;
 469         char *bufptr = buffer;
 470
 471         struct rlimit rlim[RLIM_NLIMITS];
 472
 473         if (!lock_task_sighand(task, &flags))
 474                 return 0;
 475         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 476         unlock_task_sighand(task, &flags);
 477
 478         /*
 479          * print the file header
 480          */
 481         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 482                         "Limit", "Soft Limit", "Hard Limit", "Units");
 483
 484         for (i = 0; i < RLIM_NLIMITS; i++) {
 485                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 486                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 487                                          lnames[i].name, "unlimited");
 488                 else
 489                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 490                                          lnames[i].name, rlim[i].rlim_cur);
 491
 492                 if (rlim[i].rlim_max == RLIM_INFINITY)
 493                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 494                 else
 495                         count += sprintf(&bufptr[count], "%-20lu ",
 496                                          rlim[i].rlim_max);
 497
 498                 if (lnames[i].unit)
 499                         count += sprintf(&bufptr[count], "%-10s\n",
 500                                          lnames[i].unit);
 501                 else
 502                         count += sprintf(&bufptr[count], "\n");
 503         }
 504
 505         return count;
 506 }
 507
 508 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 509 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 510 {
 511         long nr;
 512         unsigned long args[6], sp, pc;
 513         int res = lock_trace(task);
 514         if (res)
 515                 return res;
 516
 517         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 518                 res = sprintf(buffer, "running\n");
 519         else if (nr < 0)
 520                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 521         else
 522                 res = sprintf(buffer,
 523                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 524                        nr,
 525                        args[0], args[1], args[2], args[3], args[4], args[5],
 526                        sp, pc);
 527         unlock_trace(task);
 528         return res;
 529 }
 530 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 531
 532 /************************************************************************/
 533 /*                       Here the fs part begins                        */
 534 /************************************************************************/
 535
 536 /* permission checks */
 537 static int proc_fd_access_allowed(struct inode *inode)
 538 {
 539         struct task_struct *task;
 540         int allowed = 0;
 541         /* Allow access to a task's file descriptors if it is us or we
 542          * may use ptrace attach to the process and find out that
 543          * information.
 544          */
 545         task = get_proc_task(inode);
 546         if (task) {
 547                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 548                 put_task_struct(task);
 549         }
 550         return allowed;
 551 }
 552
 553 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 554 {
 555         int error;
 556         struct inode *inode = dentry->d_inode;
 557
 558         if (attr->ia_valid & ATTR_MODE)
 559                 return -EPERM;
 560
 561         error = inode_change_ok(inode, attr);
 562         if (error)
 563                 return error;
 564
 565         if ((attr->ia_valid & ATTR_SIZE) &&
 566             attr->ia_size != i_size_read(inode)) {
 567                 error = vmtruncate(inode, attr->ia_size);
 568                 if (error)
 569                         return error;
 570         }
 571
 572         setattr_copy(inode, attr);
 573         mark_inode_dirty(inode);
 574         return 0;
 575 }
 576
 577 static const struct inode_operations proc_def_inode_operations = {
 578         .setattr        = proc_setattr,
 579 };
 580
 581 static int mounts_open_common(struct inode *inode, struct file *file,
 582                               const struct seq_operations *op)
 583 {
 584         struct task_struct *task = get_proc_task(inode);
 585         struct nsproxy *nsp;
 586         struct mnt_namespace *ns = NULL;
 587         struct path root;
 588         struct proc_mounts *p;
 589         int ret = -EINVAL;
 590
 591         if (task) {
 592                 rcu_read_lock();
 593                 nsp = task_nsproxy(task);
 594                 if (nsp) {
 595                         ns = nsp->mnt_ns;
 596                         if (ns)
 597                                 get_mnt_ns(ns);
 598                 }
 599                 rcu_read_unlock();
 600                 if (ns && get_task_root(task, &root) == 0)
 601                         ret = 0;
 602                 put_task_struct(task);
 603         }
 604
 605         if (!ns)
 606                 goto err;
 607         if (ret)
 608                 goto err_put_ns;
 609
 610         ret = -ENOMEM;
 611         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 612         if (!p)
 613                 goto err_put_path;
 614
 615         file->private_data = &p->m;
 616         ret = seq_open(file, op);
 617         if (ret)
 618                 goto err_free;
 619
 620         p->m.private = p;
 621         p->ns = ns;
 622         p->root = root;
 623         p->m.poll_event = ns->event;
 624
 625         return 0;
 626
 627  err_free:
 628         kfree(p);
 629  err_put_path:
 630         path_put(&root);
 631  err_put_ns:
 632         put_mnt_ns(ns);
 633  err:
 634         return ret;
 635 }
 636
 637 static int mounts_release(struct inode *inode, struct file *file)
 638 {
 639         struct proc_mounts *p = file->private_data;
 640         path_put(&p->root);
 641         put_mnt_ns(p->ns);
 642         return seq_release(inode, file);
 643 }
 644
 645 static unsigned mounts_poll(struct file *file, poll_table *wait)
 646 {
 647         struct proc_mounts *p = file->private_data;
 648         unsigned res = POLLIN | POLLRDNORM;
 649
 650         poll_wait(file, &p->ns->poll, wait);
 651         if (mnt_had_events(p))
 652                 res |= POLLERR | POLLPRI;
 653
 654         return res;
 655 }
 656
 657 static int mounts_open(struct inode *inode, struct file *file)
 658 {
 659         return mounts_open_common(inode, file, &mounts_op);
 660 }
 661
 662 static const struct file_operations proc_mounts_operations = {
 663         .open           = mounts_open,
 664         .read           = seq_read,
 665         .llseek         = seq_lseek,
 666         .release        = mounts_release,
 667         .poll           = mounts_poll,
 668 };
 669
 670 static int mountinfo_open(struct inode *inode, struct file *file)
 671 {
 672         return mounts_open_common(inode, file, &mountinfo_op);
 673 }
 674
 675 static const struct file_operations proc_mountinfo_operations = {
 676         .open           = mountinfo_open,
 677         .read           = seq_read,
 678         .llseek         = seq_lseek,
 679         .release        = mounts_release,
 680         .poll           = mounts_poll,
 681 };
 682
 683 static int mountstats_open(struct inode *inode, struct file *file)
 684 {
 685         return mounts_open_common(inode, file, &mountstats_op);
 686 }
 687
 688 static const struct file_operations proc_mountstats_operations = {
 689         .open           = mountstats_open,
 690         .read           = seq_read,
 691         .llseek         = seq_lseek,
 692         .release        = mounts_release,
 693 };
 694
 695 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 696
 697 static ssize_t proc_info_read(struct file * file, char __user * buf,
 698                           size_t count, loff_t *ppos)
 699 {
 700         struct inode * inode = file->f_path.dentry->d_inode;
 701         unsigned long page;
 702         ssize_t length;
 703         struct task_struct *task = get_proc_task(inode);
 704
 705         length = -ESRCH;
 706         if (!task)
 707                 goto out_no_task;
 708
 709         if (count > PROC_BLOCK_SIZE)
 710                 count = PROC_BLOCK_SIZE;
 711
 712         length = -ENOMEM;
 713         if (!(page = __get_free_page(GFP_TEMPORARY)))
 714                 goto out;
 715
 716         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 717
 718         if (length >= 0)
 719                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 720         free_page(page);
 721 out:
 722         put_task_struct(task);
 723 out_no_task:
 724         return length;
 725 }
 726
 727 static const struct file_operations proc_info_file_operations = {
 728         .read           = proc_info_read,
 729         .llseek         = generic_file_llseek,
 730 };
 731
 732 static int proc_single_show(struct seq_file *m, void *v)
 733 {
 734         struct inode *inode = m->private;
 735         struct pid_namespace *ns;
 736         struct pid *pid;
 737         struct task_struct *task;
 738         int ret;
 739
 740         ns = inode->i_sb->s_fs_info;
 741         pid = proc_pid(inode);
 742         task = get_pid_task(pid, PIDTYPE_PID);
 743         if (!task)
 744                 return -ESRCH;
 745
 746         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 747
 748         put_task_struct(task);
 749         return ret;
 750 }
 751
 752 static int proc_single_open(struct inode *inode, struct file *filp)
 753 {
 754         return single_open(filp, proc_single_show, inode);
 755 }
 756
 757 static const struct file_operations proc_single_file_operations = {
 758         .open           = proc_single_open,
 759         .read           = seq_read,
 760         .llseek         = seq_lseek,
 761         .release        = single_release,
 762 };
 763
 764 static int mem_open(struct inode* inode, struct file* file)
 765 {
 766         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 767         struct mm_struct *mm;
 768
 769         if (!task)
 770                 return -ESRCH;
 771
 772         mm = mm_access(task, PTRACE_MODE_ATTACH);
 773         put_task_struct(task);
 774
 775         if (IS_ERR(mm))
 776                 return PTR_ERR(mm);
 777
 778         /* OK to pass negative loff_t, we can catch out-of-range */
 779         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 780         file->private_data = mm;
 781
 782         return 0;
 783 }
 784
 785 static ssize_t mem_read(struct file * file, char __user * buf,
 786                         size_t count, loff_t *ppos)
 787 {
 788         int ret;
 789         char *page;
 790         unsigned long src = *ppos;
 791         struct mm_struct *mm = file->private_data;
 792
 793         if (!mm)
 794                 return 0;
 795
 796         page = (char *)__get_free_page(GFP_TEMPORARY);
 797         if (!page)
 798                 return -ENOMEM;
 799
 800         ret = 0;
 801
 802         while (count > 0) {
 803                 int this_len, retval;
 804
 805                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 806                 retval = access_remote_vm(mm, src, page, this_len, 0);
 807                 if (!retval) {
 808                         if (!ret)
 809                                 ret = -EIO;
 810                         break;
 811                 }
 812
 813                 if (copy_to_user(buf, page, retval)) {
 814                         ret = -EFAULT;
 815                         break;
 816                 }
 817
 818                 ret += retval;
 819                 src += retval;
 820                 buf += retval;
 821                 count -= retval;
 822         }
 823         *ppos = src;
 824
 825         free_page((unsigned long) page);
 826         return ret;
 827 }
 828
 829 static ssize_t mem_write(struct file * file, const char __user *buf,
 830                          size_t count, loff_t *ppos)
 831 {
 832         int copied;
 833         char *page;
 834         unsigned long dst = *ppos;
 835         struct mm_struct *mm = file->private_data;
 836
 837         if (!mm)
 838                 return 0;
 839
 840         page = (char *)__get_free_page(GFP_TEMPORARY);
 841         if (!page)
 842                 return -ENOMEM;
 843
 844         copied = 0;
 845         while (count > 0) {
 846                 int this_len, retval;
 847
 848                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 849                 if (copy_from_user(page, buf, this_len)) {
 850                         copied = -EFAULT;
 851                         break;
 852                 }
 853                 retval = access_remote_vm(mm, dst, page, this_len, 1);
 854                 if (!retval) {
 855                         if (!copied)
 856                                 copied = -EIO;
 857                         break;
 858                 }
 859                 copied += retval;
 860                 buf += retval;
 861                 dst += retval;
 862                 count -= retval;
 863         }
 864         *ppos = dst;
 865
 866         free_page((unsigned long) page);
 867         return copied;
 868 }
 869
 870 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 871 {
 872         switch (orig) {
 873         case 0:
 874                 file->f_pos = offset;
 875                 break;
 876         case 1:
 877                 file->f_pos += offset;
 878                 break;
 879         default:
 880                 return -EINVAL;
 881         }
 882         force_successful_syscall_return();
 883         return file->f_pos;
 884 }
 885
 886 static int mem_release(struct inode *inode, struct file *file)
 887 {
 888         struct mm_struct *mm = file->private_data;
 889
 890         mmput(mm);
 891         return 0;
 892 }
 893
 894 static const struct file_operations proc_mem_operations = {
 895         .llseek         = mem_lseek,
 896         .read           = mem_read,
 897         .write          = mem_write,
 898         .open           = mem_open,
 899         .release        = mem_release,
 900 };
 901
 902 static ssize_t environ_read(struct file *file, char __user *buf,
 903                         size_t count, loff_t *ppos)
 904 {
 905         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 906         char *page;
 907         unsigned long src = *ppos;
 908         int ret = -ESRCH;
 909         struct mm_struct *mm;
 910
 911         if (!task)
 912                 goto out_no_task;
 913
 914         ret = -ENOMEM;
 915         page = (char *)__get_free_page(GFP_TEMPORARY);
 916         if (!page)
 917                 goto out;
 918
 919
 920         mm = mm_for_maps(task);
 921         ret = PTR_ERR(mm);
 922         if (!mm || IS_ERR(mm))
 923                 goto out_free;
 924
 925         ret = 0;
 926         while (count > 0) {
 927                 int this_len, retval, max_len;
 928
 929                 this_len = mm->env_end - (mm->env_start + src);
 930
 931                 if (this_len <= 0)
 932                         break;
 933
 934                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 935                 this_len = (this_len > max_len) ? max_len : this_len;
 936
 937                 retval = access_process_vm(task, (mm->env_start + src),
 938                         page, this_len, 0);
 939
 940                 if (retval <= 0) {
 941                         ret = retval;
 942                         break;
 943                 }
 944
 945                 if (copy_to_user(buf, page, retval)) {
 946                         ret = -EFAULT;
 947                         break;
 948                 }
 949
 950                 ret += retval;
 951                 src += retval;
 952                 buf += retval;
 953                 count -= retval;
 954         }
 955         *ppos = src;
 956
 957         mmput(mm);
 958 out_free:
 959         free_page((unsigned long) page);
 960 out:
 961         put_task_struct(task);
 962 out_no_task:
 963         return ret;
 964 }
 965
 966 static const struct file_operations proc_environ_operations = {
 967         .read           = environ_read,
 968         .llseek         = generic_file_llseek,
 969 };
 970
 971 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 972                                 size_t count, loff_t *ppos)
 973 {
 974         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 975         char buffer[PROC_NUMBUF];
 976         size_t len;
 977         int oom_adjust = OOM_DISABLE;
 978         unsigned long flags;
 979
 980         if (!task)
 981                 return -ESRCH;
 982
 983         if (lock_task_sighand(task, &flags)) {
 984                 oom_adjust = task->signal->oom_adj;
 985                 unlock_task_sighand(task, &flags);
 986         }
 987
 988         put_task_struct(task);
 989
 990         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 991
 992         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 993 }
 994
 995 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 996                                 size_t count, loff_t *ppos)
 997 {
 998         struct task_struct *task;
 999         char buffer[PROC_NUMBUF];
1000         int oom_adjust;
1001         unsigned long flags;
1002         int err;
1003
1004         memset(buffer, 0, sizeof(buffer));
1005         if (count > sizeof(buffer) - 1)
1006                 count = sizeof(buffer) - 1;
1007         if (copy_from_user(buffer, buf, count)) {
1008                 err = -EFAULT;
1009                 goto out;
1010         }
1011
1012         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
1013         if (err)
1014                 goto out;
1015         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1016              oom_adjust != OOM_DISABLE) {
1017                 err = -EINVAL;
1018                 goto out;
1019         }
1020
1021         task = get_proc_task(file->f_path.dentry->d_inode);
1022         if (!task) {
1023                 err = -ESRCH;
1024                 goto out;
1025         }
1026
1027         task_lock(task);
1028         if (!task->mm) {
1029                 err = -EINVAL;
1030                 goto err_task_lock;
1031         }
1032
1033         if (!lock_task_sighand(task, &flags)) {
1034                 err = -ESRCH;
1035                 goto err_task_lock;
1036         }
1037
1038         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1039                 err = -EACCES;
1040                 goto err_sighand;
1041         }
1042
1043         /*
1044          * Warn that /proc/pid/oom_adj is deprecated, see
1045          * Documentation/feature-removal-schedule.txt.
1046          */
1047         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1048                   current->comm, task_pid_nr(current), task_pid_nr(task),
1049                   task_pid_nr(task));
1050         task->signal->oom_adj = oom_adjust;
1051         /*
1052          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1053          * value is always attainable.
1054          */
1055         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1056                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1057         else
1058                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1059                                                                 -OOM_DISABLE;
1060 err_sighand:
1061         unlock_task_sighand(task, &flags);
1062 err_task_lock:
1063         task_unlock(task);
1064         put_task_struct(task);
1065 out:
1066         return err < 0 ? err : count;
1067 }
1068
1069 static const struct file_operations proc_oom_adjust_operations = {
1070         .read           = oom_adjust_read,
1071         .write          = oom_adjust_write,
1072         .llseek         = generic_file_llseek,
1073 };
1074
1075 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1076                                         size_t count, loff_t *ppos)
1077 {
1078         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1079         char buffer[PROC_NUMBUF];
1080         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1081         unsigned long flags;
1082         size_t len;
1083
1084         if (!task)
1085                 return -ESRCH;
1086         if (lock_task_sighand(task, &flags)) {
1087                 oom_score_adj = task->signal->oom_score_adj;
1088                 unlock_task_sighand(task, &flags);
1089         }
1090         put_task_struct(task);
1091         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1092         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1093 }
1094
1095 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1096                                         size_t count, loff_t *ppos)
1097 {
1098         struct task_struct *task;
1099         char buffer[PROC_NUMBUF];
1100         unsigned long flags;
1101         int oom_score_adj;
1102         int err;
1103
1104         memset(buffer, 0, sizeof(buffer));
1105         if (count > sizeof(buffer) - 1)
1106                 count = sizeof(buffer) - 1;
1107         if (copy_from_user(buffer, buf, count)) {
1108                 err = -EFAULT;
1109                 goto out;
1110         }
1111
1112         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1113         if (err)
1114                 goto out;
1115         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1116                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1117                 err = -EINVAL;
1118                 goto out;
1119         }
1120
1121         task = get_proc_task(file->f_path.dentry->d_inode);
1122         if (!task) {
1123                 err = -ESRCH;
1124                 goto out;
1125         }
1126
1127         task_lock(task);
1128         if (!task->mm) {
1129                 err = -EINVAL;
1130                 goto err_task_lock;
1131         }
1132
1133         if (!lock_task_sighand(task, &flags)) {
1134                 err = -ESRCH;
1135                 goto err_task_lock;
1136         }
1137
1138         if (oom_score_adj < task->signal->oom_score_adj_min &&
1139                         !capable(CAP_SYS_RESOURCE)) {
1140                 err = -EACCES;
1141                 goto err_sighand;
1142         }
1143
1144         task->signal->oom_score_adj = oom_score_adj;
1145         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1146                 task->signal->oom_score_adj_min = oom_score_adj;
1147         /*
1148          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1149          * always attainable.
1150          */
1151         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1152                 task->signal->oom_adj = OOM_DISABLE;
1153         else
1154                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1155                                                         OOM_SCORE_ADJ_MAX;
1156 err_sighand:
1157         unlock_task_sighand(task, &flags);
1158 err_task_lock:
1159         task_unlock(task);
1160         put_task_struct(task);
1161 out:
1162         return err < 0 ? err : count;
1163 }
1164
1165 static const struct file_operations proc_oom_score_adj_operations = {
1166         .read           = oom_score_adj_read,
1167         .write          = oom_score_adj_write,
1168         .llseek         = default_llseek,
1169 };
1170
1171 #ifdef CONFIG_AUDITSYSCALL
1172 #define TMPBUFLEN 21
1173 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1174                                   size_t count, loff_t *ppos)
1175 {
1176         struct inode * inode = file->f_path.dentry->d_inode;
1177         struct task_struct *task = get_proc_task(inode);
1178         ssize_t length;
1179         char tmpbuf[TMPBUFLEN];
1180
1181         if (!task)
1182                 return -ESRCH;
1183         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1184                                 audit_get_loginuid(task));
1185         put_task_struct(task);
1186         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1187 }
1188
1189 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1190                                    size_t count, loff_t *ppos)
1191 {
1192         struct inode * inode = file->f_path.dentry->d_inode;
1193         char *page, *tmp;
1194         ssize_t length;
1195         uid_t loginuid;
1196
1197         if (!capable(CAP_AUDIT_CONTROL))
1198                 return -EPERM;
1199
1200         rcu_read_lock();
1201         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1202                 rcu_read_unlock();
1203                 return -EPERM;
1204         }
1205         rcu_read_unlock();
1206
1207         if (count >= PAGE_SIZE)
1208                 count = PAGE_SIZE - 1;
1209
1210         if (*ppos != 0) {
1211                 /* No partial writes. */
1212                 return -EINVAL;
1213         }
1214         page = (char*)__get_free_page(GFP_TEMPORARY);
1215         if (!page)
1216                 return -ENOMEM;
1217         length = -EFAULT;
1218         if (copy_from_user(page, buf, count))
1219                 goto out_free_page;
1220
1221         page[count] = '\0';
1222         loginuid = simple_strtoul(page, &tmp, 10);
1223         if (tmp == page) {
1224                 length = -EINVAL;
1225                 goto out_free_page;
1226
1227         }
1228         length = audit_set_loginuid(current, loginuid);
1229         if (likely(length == 0))
1230                 length = count;
1231
1232 out_free_page:
1233         free_page((unsigned long) page);
1234         return length;
1235 }
1236
1237 static const struct file_operations proc_loginuid_operations = {
1238         .read           = proc_loginuid_read,
1239         .write          = proc_loginuid_write,
1240         .llseek         = generic_file_llseek,
1241 };
1242
1243 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1244                                   size_t count, loff_t *ppos)
1245 {
1246         struct inode * inode = file->f_path.dentry->d_inode;
1247         struct task_struct *task = get_proc_task(inode);
1248         ssize_t length;
1249         char tmpbuf[TMPBUFLEN];
1250
1251         if (!task)
1252                 return -ESRCH;
1253         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1254                                 audit_get_sessionid(task));
1255         put_task_struct(task);
1256         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1257 }
1258
1259 static const struct file_operations proc_sessionid_operations = {
1260         .read           = proc_sessionid_read,
1261         .llseek         = generic_file_llseek,
1262 };
1263 #endif
1264
1265 #ifdef CONFIG_FAULT_INJECTION
1266 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1267                                       size_t count, loff_t *ppos)
1268 {
1269         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1270         char buffer[PROC_NUMBUF];
1271         size_t len;
1272         int make_it_fail;
1273
1274         if (!task)
1275                 return -ESRCH;
1276         make_it_fail = task->make_it_fail;
1277         put_task_struct(task);
1278
1279         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1280
1281         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1282 }
1283
1284 static ssize_t proc_fault_inject_write(struct file * file,
1285                         const char __user * buf, size_t count, loff_t *ppos)
1286 {
1287         struct task_struct *task;
1288         char buffer[PROC_NUMBUF], *end;
1289         int make_it_fail;
1290
1291         if (!capable(CAP_SYS_RESOURCE))
1292                 return -EPERM;
1293         memset(buffer, 0, sizeof(buffer));
1294         if (count > sizeof(buffer) - 1)
1295                 count = sizeof(buffer) - 1;
1296         if (copy_from_user(buffer, buf, count))
1297                 return -EFAULT;
1298         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1299         if (*end)
1300                 return -EINVAL;
1301         task = get_proc_task(file->f_dentry->d_inode);
1302         if (!task)
1303                 return -ESRCH;
1304         task->make_it_fail = make_it_fail;
1305         put_task_struct(task);
1306
1307         return count;
1308 }
1309
1310 static const struct file_operations proc_fault_inject_operations = {
1311         .read           = proc_fault_inject_read,
1312         .write          = proc_fault_inject_write,
1313         .llseek         = generic_file_llseek,
1314 };
1315 #endif
1316
1317
1318 #ifdef CONFIG_SCHED_DEBUG
1319 /*
1320  * Print out various scheduling related per-task fields:
1321  */
1322 static int sched_show(struct seq_file *m, void *v)
1323 {
1324         struct inode *inode = m->private;
1325         struct task_struct *p;
1326
1327         p = get_proc_task(inode);
1328         if (!p)
1329                 return -ESRCH;
1330         proc_sched_show_task(p, m);
1331
1332         put_task_struct(p);
1333
1334         return 0;
1335 }
1336
1337 static ssize_t
1338 sched_write(struct file *file, const char __user *buf,
1339             size_t count, loff_t *offset)
1340 {
1341         struct inode *inode = file->f_path.dentry->d_inode;
1342         struct task_struct *p;
1343
1344         p = get_proc_task(inode);
1345         if (!p)
1346                 return -ESRCH;
1347         proc_sched_set_task(p);
1348
1349         put_task_struct(p);
1350
1351         return count;
1352 }
1353
1354 static int sched_open(struct inode *inode, struct file *filp)
1355 {
1356         return single_open(filp, sched_show, inode);
1357 }
1358
1359 static const struct file_operations proc_pid_sched_operations = {
1360         .open           = sched_open,
1361         .read           = seq_read,
1362         .write          = sched_write,
1363         .llseek         = seq_lseek,
1364         .release        = single_release,
1365 };
1366
1367 #endif
1368
1369 #ifdef CONFIG_SCHED_AUTOGROUP
1370 /*
1371  * Print out autogroup related information:
1372  */
1373 static int sched_autogroup_show(struct seq_file *m, void *v)
1374 {
1375         struct inode *inode = m->private;
1376         struct task_struct *p;
1377
1378         p = get_proc_task(inode);
1379         if (!p)
1380                 return -ESRCH;
1381         proc_sched_autogroup_show_task(p, m);
1382
1383         put_task_struct(p);
1384
1385         return 0;
1386 }
1387
1388 static ssize_t
1389 sched_autogroup_write(struct file *file, const char __user *buf,
1390             size_t count, loff_t *offset)
1391 {
1392         struct inode *inode = file->f_path.dentry->d_inode;
1393         struct task_struct *p;
1394         char buffer[PROC_NUMBUF];
1395         int nice;
1396         int err;
1397
1398         memset(buffer, 0, sizeof(buffer));
1399         if (count > sizeof(buffer) - 1)
1400                 count = sizeof(buffer) - 1;
1401         if (copy_from_user(buffer, buf, count))
1402                 return -EFAULT;
1403
1404         err = kstrtoint(strstrip(buffer), 0, &nice);
1405         if (err < 0)
1406                 return err;
1407
1408         p = get_proc_task(inode);
1409         if (!p)
1410                 return -ESRCH;
1411
1412         err = nice;
1413         err = proc_sched_autogroup_set_nice(p, &err);
1414         if (err)
1415                 count = err;
1416
1417         put_task_struct(p);
1418
1419         return count;
1420 }
1421
1422 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1423 {
1424         int ret;
1425
1426         ret = single_open(filp, sched_autogroup_show, NULL);
1427         if (!ret) {
1428                 struct seq_file *m = filp->private_data;
1429
1430                 m->private = inode;
1431         }
1432         return ret;
1433 }
1434
1435 static const struct file_operations proc_pid_sched_autogroup_operations = {
1436         .open           = sched_autogroup_open,
1437         .read           = seq_read,
1438         .write          = sched_autogroup_write,
1439         .llseek         = seq_lseek,
1440         .release        = single_release,
1441 };
1442
1443 #endif /* CONFIG_SCHED_AUTOGROUP */
1444
1445 static ssize_t comm_write(struct file *file, const char __user *buf,
1446                                 size_t count, loff_t *offset)
1447 {
1448         struct inode *inode = file->f_path.dentry->d_inode;
1449         struct task_struct *p;
1450         char buffer[TASK_COMM_LEN];
1451
1452         memset(buffer, 0, sizeof(buffer));
1453         if (count > sizeof(buffer) - 1)
1454                 count = sizeof(buffer) - 1;
1455         if (copy_from_user(buffer, buf, count))
1456                 return -EFAULT;
1457
1458         p = get_proc_task(inode);
1459         if (!p)
1460                 return -ESRCH;
1461
1462         if (same_thread_group(current, p))
1463                 set_task_comm(p, buffer);
1464         else
1465                 count = -EINVAL;
1466
1467         put_task_struct(p);
1468
1469         return count;
1470 }
1471
1472 static int comm_show(struct seq_file *m, void *v)
1473 {
1474         struct inode *inode = m->private;
1475         struct task_struct *p;
1476
1477         p = get_proc_task(inode);
1478         if (!p)
1479                 return -ESRCH;
1480
1481         task_lock(p);
1482         seq_printf(m, "%s\n", p->comm);
1483         task_unlock(p);
1484
1485         put_task_struct(p);
1486
1487         return 0;
1488 }
1489
1490 static int comm_open(struct inode *inode, struct file *filp)
1491 {
1492         return single_open(filp, comm_show, inode);
1493 }
1494
1495 static const struct file_operations proc_pid_set_comm_operations = {
1496         .open           = comm_open,
1497         .read           = seq_read,
1498         .write          = comm_write,
1499         .llseek         = seq_lseek,
1500         .release        = single_release,
1501 };
1502
1503 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1504 {
1505         struct task_struct *task;
1506         struct mm_struct *mm;
1507         struct file *exe_file;
1508
1509         task = get_proc_task(inode);
1510         if (!task)
1511                 return -ENOENT;
1512         mm = get_task_mm(task);
1513         put_task_struct(task);
1514         if (!mm)
1515                 return -ENOENT;
1516         exe_file = get_mm_exe_file(mm);
1517         mmput(mm);
1518         if (exe_file) {
1519                 *exe_path = exe_file->f_path;
1520                 path_get(&exe_file->f_path);
1521                 fput(exe_file);
1522                 return 0;
1523         } else
1524                 return -ENOENT;
1525 }
1526
1527 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1528 {
1529         struct inode *inode = dentry->d_inode;
1530         int error = -EACCES;
1531
1532         /* We don't need a base pointer in the /proc filesystem */
1533         path_put(&nd->path);
1534
1535         /* Are we allowed to snoop on the tasks file descriptors? */
1536         if (!proc_fd_access_allowed(inode))
1537                 goto out;
1538
1539         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1540 out:
1541         return ERR_PTR(error);
1542 }
1543
1544 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1545 {
1546         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1547         char *pathname;
1548         int len;
1549
1550         if (!tmp)
1551                 return -ENOMEM;
1552
1553         pathname = d_path(path, tmp, PAGE_SIZE);
1554         len = PTR_ERR(pathname);
1555         if (IS_ERR(pathname))
1556                 goto out;
1557         len = tmp + PAGE_SIZE - 1 - pathname;
1558
1559         if (len > buflen)
1560                 len = buflen;
1561         if (copy_to_user(buffer, pathname, len))
1562                 len = -EFAULT;
1563  out:
1564         free_page((unsigned long)tmp);
1565         return len;
1566 }
1567
1568 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1569 {
1570         int error = -EACCES;
1571         struct inode *inode = dentry->d_inode;
1572         struct path path;
1573
1574         /* Are we allowed to snoop on the tasks file descriptors? */
1575         if (!proc_fd_access_allowed(inode))
1576                 goto out;
1577
1578         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1579         if (error)
1580                 goto out;
1581
1582         error = do_proc_readlink(&path, buffer, buflen);
1583         path_put(&path);
1584 out:
1585         return error;
1586 }
1587
1588 static const struct inode_operations proc_pid_link_inode_operations = {
1589         .readlink       = proc_pid_readlink,
1590         .follow_link    = proc_pid_follow_link,
1591         .setattr        = proc_setattr,
1592 };
1593
1594
1595 /* building an inode */
1596
1597 static int task_dumpable(struct task_struct *task)
1598 {
1599         int dumpable = 0;
1600         struct mm_struct *mm;
1601
1602         task_lock(task);
1603         mm = task->mm;
1604         if (mm)
1605                 dumpable = get_dumpable(mm);
1606         task_unlock(task);
1607         if(dumpable == 1)
1608                 return 1;
1609         return 0;
1610 }
1611
1612 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1613 {
1614         struct inode * inode;
1615         struct proc_inode *ei;
1616         const struct cred *cred;
1617
1618         /* We need a new inode */
1619
1620         inode = new_inode(sb);
1621         if (!inode)
1622                 goto out;
1623
1624         /* Common stuff */
1625         ei = PROC_I(inode);
1626         inode->i_ino = get_next_ino();
1627         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1628         inode->i_op = &proc_def_inode_operations;
1629
1630         /*
1631          * grab the reference to task.
1632          */
1633         ei->pid = get_task_pid(task, PIDTYPE_PID);
1634         if (!ei->pid)
1635                 goto out_unlock;
1636
1637         if (task_dumpable(task)) {
1638                 rcu_read_lock();
1639                 cred = __task_cred(task);
1640                 inode->i_uid = cred->euid;
1641                 inode->i_gid = cred->egid;
1642                 rcu_read_unlock();
1643         }
1644         security_task_to_inode(task, inode);
1645
1646 out:
1647         return inode;
1648
1649 out_unlock:
1650         iput(inode);
1651         return NULL;
1652 }
1653
1654 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1655 {
1656         struct inode *inode = dentry->d_inode;
1657         struct task_struct *task;
1658         const struct cred *cred;
1659
1660         generic_fillattr(inode, stat);
1661
1662         rcu_read_lock();
1663         stat->uid = 0;
1664         stat->gid = 0;
1665         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1666         if (task) {
1667                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1668                     task_dumpable(task)) {
1669                         cred = __task_cred(task);
1670                         stat->uid = cred->euid;
1671                         stat->gid = cred->egid;
1672                 }
1673         }
1674         rcu_read_unlock();
1675         return 0;
1676 }
1677
1678 /* dentry stuff */
1679
1680 /*
1681  *      Exceptional case: normally we are not allowed to unhash a busy
1682  * directory. In this case, however, we can do it - no aliasing problems
1683  * due to the way we treat inodes.
1684  *
1685  * Rewrite the inode's ownerships here because the owning task may have
1686  * performed a setuid(), etc.
1687  *
1688  * Before the /proc/pid/status file was created the only way to read
1689  * the effective uid of a /process was to stat /proc/pid.  Reading
1690  * /proc/pid/status is slow enough that procps and other packages
1691  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1692  * made this apply to all per process world readable and executable
1693  * directories.
1694  */
1695 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1696 {
1697         struct inode *inode;
1698         struct task_struct *task;
1699         const struct cred *cred;
1700
1701         if (nd && nd->flags & LOOKUP_RCU)
1702                 return -ECHILD;
1703
1704         inode = dentry->d_inode;
1705         task = get_proc_task(inode);
1706
1707         if (task) {
1708                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1709                     task_dumpable(task)) {
1710                         rcu_read_lock();
1711                         cred = __task_cred(task);
1712                         inode->i_uid = cred->euid;
1713                         inode->i_gid = cred->egid;
1714                         rcu_read_unlock();
1715                 } else {
1716                         inode->i_uid = 0;
1717                         inode->i_gid = 0;
1718                 }
1719                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1720                 security_task_to_inode(task, inode);
1721                 put_task_struct(task);
1722                 return 1;
1723         }
1724         d_drop(dentry);
1725         return 0;
1726 }
1727
1728 static int pid_delete_dentry(const struct dentry * dentry)
1729 {
1730         /* Is the task we represent dead?
1731          * If so, then don't put the dentry on the lru list,
1732          * kill it immediately.
1733          */
1734         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1735 }
1736
1737 const struct dentry_operations pid_dentry_operations =
1738 {
1739         .d_revalidate   = pid_revalidate,
1740         .d_delete       = pid_delete_dentry,
1741 };
1742
1743 /* Lookups */
1744
1745 /*
1746  * Fill a directory entry.
1747  *
1748  * If possible create the dcache entry and derive our inode number and
1749  * file type from dcache entry.
1750  *
1751  * Since all of the proc inode numbers are dynamically generated, the inode
1752  * numbers do not exist until the inode is cache.  This means creating the
1753  * the dcache entry in readdir is necessary to keep the inode numbers
1754  * reported by readdir in sync with the inode numbers reported
1755  * by stat.
1756  */
1757 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1758         const char *name, int len,
1759         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1760 {
1761         struct dentry *child, *dir = filp->f_path.dentry;
1762         struct inode *inode;
1763         struct qstr qname;
1764         ino_t ino = 0;
1765         unsigned type = DT_UNKNOWN;
1766
1767         qname.name = name;
1768         qname.len  = len;
1769         qname.hash = full_name_hash(name, len);
1770
1771         child = d_lookup(dir, &qname);
1772         if (!child) {
1773                 struct dentry *new;
1774                 new = d_alloc(dir, &qname);
1775                 if (new) {
1776                         child = instantiate(dir->d_inode, new, task, ptr);
1777                         if (child)
1778                                 dput(new);
1779                         else
1780                                 child = new;
1781                 }
1782         }
1783         if (!child || IS_ERR(child) || !child->d_inode)
1784                 goto end_instantiate;
1785         inode = child->d_inode;
1786         if (inode) {
1787                 ino = inode->i_ino;
1788                 type = inode->i_mode >> 12;
1789         }
1790         dput(child);
1791 end_instantiate:
1792         if (!ino)
1793                 ino = find_inode_number(dir, &qname);
1794         if (!ino)
1795                 ino = 1;
1796         return filldir(dirent, name, len, filp->f_pos, ino, type);
1797 }
1798
1799 static unsigned name_to_int(struct dentry *dentry)
1800 {
1801         const char *name = dentry->d_name.name;
1802         int len = dentry->d_name.len;
1803         unsigned n = 0;
1804
1805         if (len > 1 && *name == '0')
1806                 goto out;
1807         while (len-- > 0) {
1808                 unsigned c = *name++ - '0';
1809                 if (c > 9)
1810                         goto out;
1811                 if (n >= (~0U-9)/10)
1812                         goto out;
1813                 n *= 10;
1814                 n += c;
1815         }
1816         return n;
1817 out:
1818         return ~0U;
1819 }
1820
1821 #define PROC_FDINFO_MAX 64
1822
1823 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1824 {
1825         struct task_struct *task = get_proc_task(inode);
1826         struct files_struct *files = NULL;
1827         struct file *file;
1828         int fd = proc_fd(inode);
1829
1830         if (task) {
1831                 files = get_files_struct(task);
1832                 put_task_struct(task);
1833         }
1834         if (files) {
1835                 /*
1836                  * We are not taking a ref to the file structure, so we must
1837                  * hold ->file_lock.
1838                  */
1839                 spin_lock(&files->file_lock);
1840                 file = fcheck_files(files, fd);
1841                 if (file) {
1842                         unsigned int f_flags;
1843                         struct fdtable *fdt;
1844
1845                         fdt = files_fdtable(files);
1846                         f_flags = file->f_flags & ~O_CLOEXEC;
1847                         if (FD_ISSET(fd, fdt->close_on_exec))
1848                                 f_flags |= O_CLOEXEC;
1849
1850                         if (path) {
1851                                 *path = file->f_path;
1852                                 path_get(&file->f_path);
1853                         }
1854                         if (info)
1855                                 snprintf(info, PROC_FDINFO_MAX,
1856                                          "pos:\t%lli\n"
1857                                          "flags:\t0%o\n",
1858                                          (long long) file->f_pos,
1859                                          f_flags);
1860                         spin_unlock(&files->file_lock);
1861                         put_files_struct(files);
1862                         return 0;
1863                 }
1864                 spin_unlock(&files->file_lock);
1865                 put_files_struct(files);
1866         }
1867         return -ENOENT;
1868 }
1869
1870 static int proc_fd_link(struct inode *inode, struct path *path)
1871 {
1872         return proc_fd_info(inode, path, NULL);
1873 }
1874
1875 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1876 {
1877         struct inode *inode;
1878         struct task_struct *task;
1879         int fd;
1880         struct files_struct *files;
1881         const struct cred *cred;
1882
1883         if (nd && nd->flags & LOOKUP_RCU)
1884                 return -ECHILD;
1885
1886         inode = dentry->d_inode;
1887         task = get_proc_task(inode);
1888         fd = proc_fd(inode);
1889
1890         if (task) {
1891                 files = get_files_struct(task);
1892                 if (files) {
1893                         rcu_read_lock();
1894                         if (fcheck_files(files, fd)) {
1895                                 rcu_read_unlock();
1896                                 put_files_struct(files);
1897                                 if (task_dumpable(task)) {
1898                                         rcu_read_lock();
1899                                         cred = __task_cred(task);
1900                                         inode->i_uid = cred->euid;
1901                                         inode->i_gid = cred->egid;
1902                                         rcu_read_unlock();
1903                                 } else {
1904                                         inode->i_uid = 0;
1905                                         inode->i_gid = 0;
1906                                 }
1907                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1908                                 security_task_to_inode(task, inode);
1909                                 put_task_struct(task);
1910                                 return 1;
1911                         }
1912                         rcu_read_unlock();
1913                         put_files_struct(files);
1914                 }
1915                 put_task_struct(task);
1916         }
1917         d_drop(dentry);
1918         return 0;
1919 }
1920
1921 static const struct dentry_operations tid_fd_dentry_operations =
1922 {
1923         .d_revalidate   = tid_fd_revalidate,
1924         .d_delete       = pid_delete_dentry,
1925 };
1926
1927 static struct dentry *proc_fd_instantiate(struct inode *dir,
1928         struct dentry *dentry, struct task_struct *task, const void *ptr)
1929 {
1930         unsigned fd = *(const unsigned *)ptr;
1931         struct file *file;
1932         struct files_struct *files;
1933         struct inode *inode;
1934         struct proc_inode *ei;
1935         struct dentry *error = ERR_PTR(-ENOENT);
1936
1937         inode = proc_pid_make_inode(dir->i_sb, task);
1938         if (!inode)
1939                 goto out;
1940         ei = PROC_I(inode);
1941         ei->fd = fd;
1942         files = get_files_struct(task);
1943         if (!files)
1944                 goto out_iput;
1945         inode->i_mode = S_IFLNK;
1946
1947         /*
1948          * We are not taking a ref to the file structure, so we must
1949          * hold ->file_lock.
1950          */
1951         spin_lock(&files->file_lock);
1952         file = fcheck_files(files, fd);
1953         if (!file)
1954                 goto out_unlock;
1955         if (file->f_mode & FMODE_READ)
1956                 inode->i_mode |= S_IRUSR | S_IXUSR;
1957         if (file->f_mode & FMODE_WRITE)
1958                 inode->i_mode |= S_IWUSR | S_IXUSR;
1959         spin_unlock(&files->file_lock);
1960         put_files_struct(files);
1961
1962         inode->i_op = &proc_pid_link_inode_operations;
1963         inode->i_size = 64;
1964         ei->op.proc_get_link = proc_fd_link;
1965         d_set_d_op(dentry, &tid_fd_dentry_operations);
1966         d_add(dentry, inode);
1967         /* Close the race of the process dying before we return the dentry */
1968         if (tid_fd_revalidate(dentry, NULL))
1969                 error = NULL;
1970
1971  out:
1972         return error;
1973 out_unlock:
1974         spin_unlock(&files->file_lock);
1975         put_files_struct(files);
1976 out_iput:
1977         iput(inode);
1978         goto out;
1979 }
1980
1981 static struct dentry *proc_lookupfd_common(struct inode *dir,
1982                                            struct dentry *dentry,
1983                                            instantiate_t instantiate)
1984 {
1985         struct task_struct *task = get_proc_task(dir);
1986         unsigned fd = name_to_int(dentry);
1987         struct dentry *result = ERR_PTR(-ENOENT);
1988
1989         if (!task)
1990                 goto out_no_task;
1991         if (fd == ~0U)
1992                 goto out;
1993
1994         result = instantiate(dir, dentry, task, &fd);
1995 out:
1996         put_task_struct(task);
1997 out_no_task:
1998         return result;
1999 }
2000
2001 static int proc_readfd_common(struct file * filp, void * dirent,
2002                               filldir_t filldir, instantiate_t instantiate)
2003 {
2004         struct dentry *dentry = filp->f_path.dentry;
2005         struct inode *inode = dentry->d_inode;
2006         struct task_struct *p = get_proc_task(inode);
2007         unsigned int fd, ino;
2008         int retval;
2009         struct files_struct * files;
2010
2011         retval = -ENOENT;
2012         if (!p)
2013                 goto out_no_task;
2014         retval = 0;
2015
2016         fd = filp->f_pos;
2017         switch (fd) {
2018                 case 0:
2019                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2020                                 goto out;
2021                         filp->f_pos++;
2022                 case 1:
2023                         ino = parent_ino(dentry);
2024                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2025                                 goto out;
2026                         filp->f_pos++;
2027                 default:
2028                         files = get_files_struct(p);
2029                         if (!files)
2030                                 goto out;
2031                         rcu_read_lock();
2032                         for (fd = filp->f_pos-2;
2033                              fd < files_fdtable(files)->max_fds;
2034                              fd++, filp->f_pos++) {
2035                                 char name[PROC_NUMBUF];
2036                                 int len;
2037
2038                                 if (!fcheck_files(files, fd))
2039                                         continue;
2040                                 rcu_read_unlock();
2041
2042                                 len = snprintf(name, sizeof(name), "%d", fd);
2043                                 if (proc_fill_cache(filp, dirent, filldir,
2044                                                     name, len, instantiate,
2045                                                     p, &fd) < 0) {
2046                                         rcu_read_lock();
2047                                         break;
2048                                 }
2049                                 rcu_read_lock();
2050                         }
2051                         rcu_read_unlock();
2052                         put_files_struct(files);
2053         }
2054 out:
2055         put_task_struct(p);
2056 out_no_task:
2057         return retval;
2058 }
2059
2060 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2061                                     struct nameidata *nd)
2062 {
2063         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2064 }
2065
2066 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2067 {
2068         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2069 }
2070
2071 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2072                                       size_t len, loff_t *ppos)
2073 {
2074         char tmp[PROC_FDINFO_MAX];
2075         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2076         if (!err)
2077                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2078         return err;
2079 }
2080
2081 static const struct file_operations proc_fdinfo_file_operations = {
2082         .open           = nonseekable_open,
2083         .read           = proc_fdinfo_read,
2084         .llseek         = no_llseek,
2085 };
2086
2087 static const struct file_operations proc_fd_operations = {
2088         .read           = generic_read_dir,
2089         .readdir        = proc_readfd,
2090         .llseek         = default_llseek,
2091 };
2092
2093 /*
2094  * /proc/pid/fd needs a special permission handler so that a process can still
2095  * access /proc/self/fd after it has executed a setuid().
2096  */
2097 static int proc_fd_permission(struct inode *inode, int mask)
2098 {
2099         int rv = generic_permission(inode, mask);
2100         if (rv == 0)
2101                 return 0;
2102         if (task_pid(current) == proc_pid(inode))
2103                 rv = 0;
2104         return rv;
2105 }
2106
2107 /*
2108  * proc directories can do almost nothing..
2109  */
2110 static const struct inode_operations proc_fd_inode_operations = {
2111         .lookup         = proc_lookupfd,
2112         .permission     = proc_fd_permission,
2113         .setattr        = proc_setattr,
2114 };
2115
2116 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2117         struct dentry *dentry, struct task_struct *task, const void *ptr)
2118 {
2119         unsigned fd = *(unsigned *)ptr;
2120         struct inode *inode;
2121         struct proc_inode *ei;
2122         struct dentry *error = ERR_PTR(-ENOENT);
2123
2124         inode = proc_pid_make_inode(dir->i_sb, task);
2125         if (!inode)
2126                 goto out;
2127         ei = PROC_I(inode);
2128         ei->fd = fd;
2129         inode->i_mode = S_IFREG | S_IRUSR;
2130         inode->i_fop = &proc_fdinfo_file_operations;
2131         d_set_d_op(dentry, &tid_fd_dentry_operations);
2132         d_add(dentry, inode);
2133         /* Close the race of the process dying before we return the dentry */
2134         if (tid_fd_revalidate(dentry, NULL))
2135                 error = NULL;
2136
2137  out:
2138         return error;
2139 }
2140
2141 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2142                                         struct dentry *dentry,
2143                                         struct nameidata *nd)
2144 {
2145         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2146 }
2147
2148 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2149 {
2150         return proc_readfd_common(filp, dirent, filldir,
2151                                   proc_fdinfo_instantiate);
2152 }
2153
2154 static const struct file_operations proc_fdinfo_operations = {
2155         .read           = generic_read_dir,
2156         .readdir        = proc_readfdinfo,
2157         .llseek         = default_llseek,
2158 };
2159
2160 /*
2161  * proc directories can do almost nothing..
2162  */
2163 static const struct inode_operations proc_fdinfo_inode_operations = {
2164         .lookup         = proc_lookupfdinfo,
2165         .setattr        = proc_setattr,
2166 };
2167
2168
2169 static struct dentry *proc_pident_instantiate(struct inode *dir,
2170         struct dentry *dentry, struct task_struct *task, const void *ptr)
2171 {
2172         const struct pid_entry *p = ptr;
2173         struct inode *inode;
2174         struct proc_inode *ei;
2175         struct dentry *error = ERR_PTR(-ENOENT);
2176
2177         inode = proc_pid_make_inode(dir->i_sb, task);
2178         if (!inode)
2179                 goto out;
2180
2181         ei = PROC_I(inode);
2182         inode->i_mode = p->mode;
2183         if (S_ISDIR(inode->i_mode))
2184                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2185         if (p->iop)
2186                 inode->i_op = p->iop;
2187         if (p->fop)
2188                 inode->i_fop = p->fop;
2189         ei->op = p->op;
2190         d_set_d_op(dentry, &pid_dentry_operations);
2191         d_add(dentry, inode);
2192         /* Close the race of the process dying before we return the dentry */
2193         if (pid_revalidate(dentry, NULL))
2194                 error = NULL;
2195 out:
2196         return error;
2197 }
2198
2199 static struct dentry *proc_pident_lookup(struct inode *dir,
2200                                          struct dentry *dentry,
2201                                          const struct pid_entry *ents,
2202                                          unsigned int nents)
2203 {
2204         struct dentry *error;
2205         struct task_struct *task = get_proc_task(dir);
2206         const struct pid_entry *p, *last;
2207
2208         error = ERR_PTR(-ENOENT);
2209
2210         if (!task)
2211                 goto out_no_task;
2212
2213         /*
2214          * Yes, it does not scale. And it should not. Don't add
2215          * new entries into /proc/<tgid>/ without very good reasons.
2216          */
2217         last = &ents[nents - 1];
2218         for (p = ents; p <= last; p++) {
2219                 if (p->len != dentry->d_name.len)
2220                         continue;
2221                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2222                         break;
2223         }
2224         if (p > last)
2225                 goto out;
2226
2227         error = proc_pident_instantiate(dir, dentry, task, p);
2228 out:
2229         put_task_struct(task);
2230 out_no_task:
2231         return error;
2232 }
2233
2234 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2235         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2236 {
2237         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2238                                 proc_pident_instantiate, task, p);
2239 }
2240
2241 static int proc_pident_readdir(struct file *filp,
2242                 void *dirent, filldir_t filldir,
2243                 const struct pid_entry *ents, unsigned int nents)
2244 {
2245         int i;
2246         struct dentry *dentry = filp->f_path.dentry;
2247         struct inode *inode = dentry->d_inode;
2248         struct task_struct *task = get_proc_task(inode);
2249         const struct pid_entry *p, *last;
2250         ino_t ino;
2251         int ret;
2252
2253         ret = -ENOENT;
2254         if (!task)
2255                 goto out_no_task;
2256
2257         ret = 0;
2258         i = filp->f_pos;
2259         switch (i) {
2260         case 0:
2261                 ino = inode->i_ino;
2262                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2263                         goto out;
2264                 i++;
2265                 filp->f_pos++;
2266                 /* fall through */
2267         case 1:
2268                 ino = parent_ino(dentry);
2269                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2270                         goto out;
2271                 i++;
2272                 filp->f_pos++;
2273                 /* fall through */
2274         default:
2275                 i -= 2;
2276                 if (i >= nents) {
2277                         ret = 1;
2278                         goto out;
2279                 }
2280                 p = ents + i;
2281                 last = &ents[nents - 1];
2282                 while (p <= last) {
2283                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2284                                 goto out;
2285                         filp->f_pos++;
2286                         p++;
2287                 }
2288         }
2289
2290         ret = 1;
2291 out:
2292         put_task_struct(task);
2293 out_no_task:
2294         return ret;
2295 }
2296
2297 #ifdef CONFIG_SECURITY
2298 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2299                                   size_t count, loff_t *ppos)
2300 {
2301         struct inode * inode = file->f_path.dentry->d_inode;
2302         char *p = NULL;
2303         ssize_t length;
2304         struct task_struct *task = get_proc_task(inode);
2305
2306         if (!task)
2307                 return -ESRCH;
2308
2309         length = security_getprocattr(task,
2310                                       (char*)file->f_path.dentry->d_name.name,
2311                                       &p);
2312         put_task_struct(task);
2313         if (length > 0)
2314                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2315         kfree(p);
2316         return length;
2317 }
2318
2319 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2320                                    size_t count, loff_t *ppos)
2321 {
2322         struct inode * inode = file->f_path.dentry->d_inode;
2323         char *page;
2324         ssize_t length;
2325         struct task_struct *task = get_proc_task(inode);
2326
2327         length = -ESRCH;
2328         if (!task)
2329                 goto out_no_task;
2330         if (count > PAGE_SIZE)
2331                 count = PAGE_SIZE;
2332
2333         /* No partial writes. */
2334         length = -EINVAL;
2335         if (*ppos != 0)
2336                 goto out;
2337
2338         length = -ENOMEM;
2339         page = (char*)__get_free_page(GFP_TEMPORARY);
2340         if (!page)
2341                 goto out;
2342
2343         length = -EFAULT;
2344         if (copy_from_user(page, buf, count))
2345                 goto out_free;
2346
2347         /* Guard against adverse ptrace interaction */
2348         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2349         if (length < 0)
2350                 goto out_free;
2351
2352         length = security_setprocattr(task,
2353                                       (char*)file->f_path.dentry->d_name.name,
2354                                       (void*)page, count);
2355         mutex_unlock(&task->signal->cred_guard_mutex);
2356 out_free:
2357         free_page((unsigned long) page);
2358 out:
2359         put_task_struct(task);
2360 out_no_task:
2361         return length;
2362 }
2363
2364 static const struct file_operations proc_pid_attr_operations = {
2365         .read           = proc_pid_attr_read,
2366         .write          = proc_pid_attr_write,
2367         .llseek         = generic_file_llseek,
2368 };
2369
2370 static const struct pid_entry attr_dir_stuff[] = {
2371         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2372         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2373         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2374         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2375         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2376         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2377 };
2378
2379 static int proc_attr_dir_readdir(struct file * filp,
2380                              void * dirent, filldir_t filldir)
2381 {
2382         return proc_pident_readdir(filp,dirent,filldir,
2383                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2384 }
2385
2386 static const struct file_operations proc_attr_dir_operations = {
2387         .read           = generic_read_dir,
2388         .readdir        = proc_attr_dir_readdir,
2389         .llseek         = default_llseek,
2390 };
2391
2392 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2393                                 struct dentry *dentry, struct nameidata *nd)
2394 {
2395         return proc_pident_lookup(dir, dentry,
2396                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2397 }
2398
2399 static const struct inode_operations proc_attr_dir_inode_operations = {
2400         .lookup         = proc_attr_dir_lookup,
2401         .getattr        = pid_getattr,
2402         .setattr        = proc_setattr,
2403 };
2404
2405 #endif
2406
2407 #ifdef CONFIG_ELF_CORE
2408 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2409                                          size_t count, loff_t *ppos)
2410 {
2411         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2412         struct mm_struct *mm;
2413         char buffer[PROC_NUMBUF];
2414         size_t len;
2415         int ret;
2416
2417         if (!task)
2418                 return -ESRCH;
2419
2420         ret = 0;
2421         mm = get_task_mm(task);
2422         if (mm) {
2423                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2424                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2425                                 MMF_DUMP_FILTER_SHIFT));
2426                 mmput(mm);
2427                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2428         }
2429
2430         put_task_struct(task);
2431
2432         return ret;
2433 }
2434
2435 static ssize_t proc_coredump_filter_write(struct file *file,
2436                                           const char __user *buf,
2437                                           size_t count,
2438                                           loff_t *ppos)
2439 {
2440         struct task_struct *task;
2441         struct mm_struct *mm;
2442         char buffer[PROC_NUMBUF], *end;
2443         unsigned int val;
2444         int ret;
2445         int i;
2446         unsigned long mask;
2447
2448         ret = -EFAULT;
2449         memset(buffer, 0, sizeof(buffer));
2450         if (count > sizeof(buffer) - 1)
2451                 count = sizeof(buffer) - 1;
2452         if (copy_from_user(buffer, buf, count))
2453                 goto out_no_task;
2454
2455         ret = -EINVAL;
2456         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2457         if (*end == '\n')
2458                 end++;
2459         if (end - buffer == 0)
2460                 goto out_no_task;
2461
2462         ret = -ESRCH;
2463         task = get_proc_task(file->f_dentry->d_inode);
2464         if (!task)
2465                 goto out_no_task;
2466
2467         ret = end - buffer;
2468         mm = get_task_mm(task);
2469         if (!mm)
2470                 goto out_no_mm;
2471
2472         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2473                 if (val & mask)
2474                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2475                 else
2476                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2477         }
2478
2479         mmput(mm);
2480  out_no_mm:
2481         put_task_struct(task);
2482  out_no_task:
2483         return ret;
2484 }
2485
2486 static const struct file_operations proc_coredump_filter_operations = {
2487         .read           = proc_coredump_filter_read,
2488         .write          = proc_coredump_filter_write,
2489         .llseek         = generic_file_llseek,
2490 };
2491 #endif
2492
2493 /*
2494  * /proc/self:
2495  */
2496 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2497                               int buflen)
2498 {
2499         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2500         pid_t tgid = task_tgid_nr_ns(current, ns);
2501         char tmp[PROC_NUMBUF];
2502         if (!tgid)
2503                 return -ENOENT;
2504         sprintf(tmp, "%d", tgid);
2505         return vfs_readlink(dentry,buffer,buflen,tmp);
2506 }
2507
2508 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2509 {
2510         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2511         pid_t tgid = task_tgid_nr_ns(current, ns);
2512         char *name = ERR_PTR(-ENOENT);
2513         if (tgid) {
2514                 name = __getname();
2515                 if (!name)
2516                         name = ERR_PTR(-ENOMEM);
2517                 else
2518                         sprintf(name, "%d", tgid);
2519         }
2520         nd_set_link(nd, name);
2521         return NULL;
2522 }
2523
2524 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2525                                 void *cookie)
2526 {
2527         char *s = nd_get_link(nd);
2528         if (!IS_ERR(s))
2529                 __putname(s);
2530 }
2531
2532 static const struct inode_operations proc_self_inode_operations = {
2533         .readlink       = proc_self_readlink,
2534         .follow_link    = proc_self_follow_link,
2535         .put_link       = proc_self_put_link,
2536 };
2537
2538 /*
2539  * proc base
2540  *
2541  * These are the directory entries in the root directory of /proc
2542  * that properly belong to the /proc filesystem, as they describe
2543  * describe something that is process related.
2544  */
2545 static const struct pid_entry proc_base_stuff[] = {
2546         NOD("self", S_IFLNK|S_IRWXUGO,
2547                 &proc_self_inode_operations, NULL, {}),
2548 };
2549
2550 static struct dentry *proc_base_instantiate(struct inode *dir,
2551         struct dentry *dentry, struct task_struct *task, const void *ptr)
2552 {
2553         const struct pid_entry *p = ptr;
2554         struct inode *inode;
2555         struct proc_inode *ei;
2556         struct dentry *error;
2557
2558         /* Allocate the inode */
2559         error = ERR_PTR(-ENOMEM);
2560         inode = new_inode(dir->i_sb);
2561         if (!inode)
2562                 goto out;
2563
2564         /* Initialize the inode */
2565         ei = PROC_I(inode);
2566         inode->i_ino = get_next_ino();
2567         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2568
2569         /*
2570          * grab the reference to the task.
2571          */
2572         ei->pid = get_task_pid(task, PIDTYPE_PID);
2573         if (!ei->pid)
2574                 goto out_iput;
2575
2576         inode->i_mode = p->mode;
2577         if (S_ISDIR(inode->i_mode))
2578                 set_nlink(inode, 2);
2579         if (S_ISLNK(inode->i_mode))
2580                 inode->i_size = 64;
2581         if (p->iop)
2582                 inode->i_op = p->iop;
2583         if (p->fop)
2584                 inode->i_fop = p->fop;
2585         ei->op = p->op;
2586         d_add(dentry, inode);
2587         error = NULL;
2588 out:
2589         return error;
2590 out_iput:
2591         iput(inode);
2592         goto out;
2593 }
2594
2595 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2596 {
2597         struct dentry *error;
2598         struct task_struct *task = get_proc_task(dir);
2599         const struct pid_entry *p, *last;
2600
2601         error = ERR_PTR(-ENOENT);
2602
2603         if (!task)
2604                 goto out_no_task;
2605
2606         /* Lookup the directory entry */
2607         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2608         for (p = proc_base_stuff; p <= last; p++) {
2609                 if (p->len != dentry->d_name.len)
2610                         continue;
2611                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2612                         break;
2613         }
2614         if (p > last)
2615                 goto out;
2616
2617         error = proc_base_instantiate(dir, dentry, task, p);
2618
2619 out:
2620         put_task_struct(task);
2621 out_no_task:
2622         return error;
2623 }
2624
2625 static int proc_base_fill_cache(struct file *filp, void *dirent,
2626         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2627 {
2628         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2629                                 proc_base_instantiate, task, p);
2630 }
2631
2632 #ifdef CONFIG_TASK_IO_ACCOUNTING
2633 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2634 {
2635         struct task_io_accounting acct = task->ioac;
2636         unsigned long flags;
2637         int result;
2638
2639         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2640         if (result)
2641                 return result;
2642
2643         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2644                 result = -EACCES;
2645                 goto out_unlock;
2646         }
2647
2648         if (whole && lock_task_sighand(task, &flags)) {
2649                 struct task_struct *t = task;
2650
2651                 task_io_accounting_add(&acct, &task->signal->ioac);
2652                 while_each_thread(task, t)
2653                         task_io_accounting_add(&acct, &t->ioac);
2654
2655                 unlock_task_sighand(task, &flags);
2656         }
2657         result = sprintf(buffer,
2658                         "rchar: %llu\n"
2659                         "wchar: %llu\n"
2660                         "syscr: %llu\n"
2661                         "syscw: %llu\n"
2662                         "read_bytes: %llu\n"
2663                         "write_bytes: %llu\n"
2664                         "cancelled_write_bytes: %llu\n",
2665                         (unsigned long long)acct.rchar,
2666                         (unsigned long long)acct.wchar,
2667                         (unsigned long long)acct.syscr,
2668                         (unsigned long long)acct.syscw,
2669                         (unsigned long long)acct.read_bytes,
2670                         (unsigned long long)acct.write_bytes,
2671                         (unsigned long long)acct.cancelled_write_bytes);
2672 out_unlock:
2673         mutex_unlock(&task->signal->cred_guard_mutex);
2674         return result;
2675 }
2676
2677 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2678 {
2679         return do_io_accounting(task, buffer, 0);
2680 }
2681
2682 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2683 {
2684         return do_io_accounting(task, buffer, 1);
2685 }
2686 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2687
2688 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2689                                 struct pid *pid, struct task_struct *task)
2690 {
2691         int err = lock_trace(task);
2692         if (!err) {
2693                 seq_printf(m, "%08x\n", task->personality);
2694                 unlock_trace(task);
2695         }
2696         return err;
2697 }
2698
2699 /*
2700  * Thread groups
2701  */
2702 static const struct file_operations proc_task_operations;
2703 static const struct inode_operations proc_task_inode_operations;
2704
2705 static const struct pid_entry tgid_base_stuff[] = {
2706         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2707         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2708         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2709         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2710 #ifdef CONFIG_NET
2711         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2712 #endif
2713         REG("environ",    S_IRUSR, proc_environ_operations),
2714         INF("auxv",       S_IRUSR, proc_pid_auxv),
2715         ONE("status",     S_IRUGO, proc_pid_status),
2716         ONE("personality", S_IRUGO, proc_pid_personality),
2717         INF("limits",     S_IRUGO, proc_pid_limits),
2718 #ifdef CONFIG_SCHED_DEBUG
2719         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2720 #endif
2721 #ifdef CONFIG_SCHED_AUTOGROUP
2722         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2723 #endif
2724         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2725 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2726         INF("syscall",    S_IRUGO, proc_pid_syscall),
2727 #endif
2728         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2729         ONE("stat",       S_IRUGO, proc_tgid_stat),
2730         ONE("statm",      S_IRUGO, proc_pid_statm),
2731         REG("maps",       S_IRUGO, proc_maps_operations),
2732 #ifdef CONFIG_NUMA
2733         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2734 #endif
2735         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2736         LNK("cwd",        proc_cwd_link),
2737         LNK("root",       proc_root_link),
2738         LNK("exe",        proc_exe_link),
2739         REG("mounts",     S_IRUGO, proc_mounts_operations),
2740         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2741         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2742 #ifdef CONFIG_PROC_PAGE_MONITOR
2743         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2744         REG("smaps",      S_IRUGO, proc_smaps_operations),
2745         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2746 #endif
2747 #ifdef CONFIG_SECURITY
2748         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2749 #endif
2750 #ifdef CONFIG_KALLSYMS
2751         INF("wchan",      S_IRUGO, proc_pid_wchan),
2752 #endif
2753 #ifdef CONFIG_STACKTRACE
2754         ONE("stack",      S_IRUGO, proc_pid_stack),
2755 #endif
2756 #ifdef CONFIG_SCHEDSTATS
2757         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2758 #endif
2759 #ifdef CONFIG_LATENCYTOP
2760         REG("latency",  S_IRUGO, proc_lstats_operations),
2761 #endif
2762 #ifdef CONFIG_PROC_PID_CPUSET
2763         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2764 #endif
2765 #ifdef CONFIG_CGROUPS
2766         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2767 #endif
2768         INF("oom_score",  S_IRUGO, proc_oom_score),
2769         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2770         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2771 #ifdef CONFIG_AUDITSYSCALL
2772         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2773         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2774 #endif
2775 #ifdef CONFIG_FAULT_INJECTION
2776         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2777 #endif
2778 #ifdef CONFIG_ELF_CORE
2779         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2780 #endif
2781 #ifdef CONFIG_TASK_IO_ACCOUNTING
2782         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2783 #endif
2784 #ifdef CONFIG_HARDWALL
2785         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2786 #endif
2787 };
2788
2789 static int proc_tgid_base_readdir(struct file * filp,
2790                              void * dirent, filldir_t filldir)
2791 {
2792         return proc_pident_readdir(filp,dirent,filldir,
2793                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2794 }
2795
2796 static const struct file_operations proc_tgid_base_operations = {
2797         .read           = generic_read_dir,
2798         .readdir        = proc_tgid_base_readdir,
2799         .llseek         = default_llseek,
2800 };
2801
2802 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2803         return proc_pident_lookup(dir, dentry,
2804                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2805 }
2806
2807 static const struct inode_operations proc_tgid_base_inode_operations = {
2808         .lookup         = proc_tgid_base_lookup,
2809         .getattr        = pid_getattr,
2810         .setattr        = proc_setattr,
2811 };
2812
2813 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2814 {
2815         struct dentry *dentry, *leader, *dir;
2816         char buf[PROC_NUMBUF];
2817         struct qstr name;
2818
2819         name.name = buf;
2820         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2821         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2822         if (dentry) {
2823                 shrink_dcache_parent(dentry);
2824                 d_drop(dentry);
2825                 dput(dentry);
2826         }
2827
2828         name.name = buf;
2829         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2830         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2831         if (!leader)
2832                 goto out;
2833
2834         name.name = "task";
2835         name.len = strlen(name.name);
2836         dir = d_hash_and_lookup(leader, &name);
2837         if (!dir)
2838                 goto out_put_leader;
2839
2840         name.name = buf;
2841         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2842         dentry = d_hash_and_lookup(dir, &name);
2843         if (dentry) {
2844                 shrink_dcache_parent(dentry);
2845                 d_drop(dentry);
2846                 dput(dentry);
2847         }
2848
2849         dput(dir);
2850 out_put_leader:
2851         dput(leader);
2852 out:
2853         return;
2854 }
2855
2856 /**
2857  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2858  * @task: task that should be flushed.
2859  *
2860  * When flushing dentries from proc, one needs to flush them from global
2861  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2862  * in. This call is supposed to do all of this job.
2863  *
2864  * Looks in the dcache for
2865  * /proc/@pid
2866  * /proc/@tgid/task/@pid
2867  * if either directory is present flushes it and all of it'ts children
2868  * from the dcache.
2869  *
2870  * It is safe and reasonable to cache /proc entries for a task until
2871  * that task exits.  After that they just clog up the dcache with
2872  * useless entries, possibly causing useful dcache entries to be
2873  * flushed instead.  This routine is proved to flush those useless
2874  * dcache entries at process exit time.
2875  *
2876  * NOTE: This routine is just an optimization so it does not guarantee
2877  *       that no dcache entries will exist at process exit time it
2878  *       just makes it very unlikely that any will persist.
2879  */
2880
2881 void proc_flush_task(struct task_struct *task)
2882 {
2883         int i;
2884         struct pid *pid, *tgid;
2885         struct upid *upid;
2886
2887         pid = task_pid(task);
2888         tgid = task_tgid(task);
2889
2890         for (i = 0; i <= pid->level; i++) {
2891                 upid = &pid->numbers[i];
2892                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2893                                         tgid->numbers[i].nr);
2894         }
2895
2896         upid = &pid->numbers[pid->level];
2897         if (upid->nr == 1)
2898                 pid_ns_release_proc(upid->ns);
2899 }
2900
2901 static struct dentry *proc_pid_instantiate(struct inode *dir,
2902                                            struct dentry * dentry,
2903                                            struct task_struct *task, const void *ptr)
2904 {
2905         struct dentry *error = ERR_PTR(-ENOENT);
2906         struct inode *inode;
2907
2908         inode = proc_pid_make_inode(dir->i_sb, task);
2909         if (!inode)
2910                 goto out;
2911
2912         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2913         inode->i_op = &proc_tgid_base_inode_operations;
2914         inode->i_fop = &proc_tgid_base_operations;
2915         inode->i_flags|=S_IMMUTABLE;
2916
2917         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2918                                                   ARRAY_SIZE(tgid_base_stuff)));
2919
2920         d_set_d_op(dentry, &pid_dentry_operations);
2921
2922         d_add(dentry, inode);
2923         /* Close the race of the process dying before we return the dentry */
2924         if (pid_revalidate(dentry, NULL))
2925                 error = NULL;
2926 out:
2927         return error;
2928 }
2929
2930 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2931 {
2932         struct dentry *result;
2933         struct task_struct *task;
2934         unsigned tgid;
2935         struct pid_namespace *ns;
2936
2937         result = proc_base_lookup(dir, dentry);
2938         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2939                 goto out;
2940
2941         tgid = name_to_int(dentry);
2942         if (tgid == ~0U)
2943                 goto out;
2944
2945         ns = dentry->d_sb->s_fs_info;
2946         rcu_read_lock();
2947         task = find_task_by_pid_ns(tgid, ns);
2948         if (task)
2949                 get_task_struct(task);
2950         rcu_read_unlock();
2951         if (!task)
2952                 goto out;
2953
2954         result = proc_pid_instantiate(dir, dentry, task, NULL);
2955         put_task_struct(task);
2956 out:
2957         return result;
2958 }
2959
2960 /*
2961  * Find the first task with tgid >= tgid
2962  *
2963  */
2964 struct tgid_iter {
2965         unsigned int tgid;
2966         struct task_struct *task;
2967 };
2968 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2969 {
2970         struct pid *pid;
2971
2972         if (iter.task)
2973                 put_task_struct(iter.task);
2974         rcu_read_lock();
2975 retry:
2976         iter.task = NULL;
2977         pid = find_ge_pid(iter.tgid, ns);
2978         if (pid) {
2979                 iter.tgid = pid_nr_ns(pid, ns);
2980                 iter.task = pid_task(pid, PIDTYPE_PID);
2981                 /* What we to know is if the pid we have find is the
2982                  * pid of a thread_group_leader.  Testing for task
2983                  * being a thread_group_leader is the obvious thing
2984                  * todo but there is a window when it fails, due to
2985                  * the pid transfer logic in de_thread.
2986                  *
2987                  * So we perform the straight forward test of seeing
2988                  * if the pid we have found is the pid of a thread
2989                  * group leader, and don't worry if the task we have
2990                  * found doesn't happen to be a thread group leader.
2991                  * As we don't care in the case of readdir.
2992                  */
2993                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2994                         iter.tgid += 1;
2995                         goto retry;
2996                 }
2997                 get_task_struct(iter.task);
2998         }
2999         rcu_read_unlock();
3000         return iter;
3001 }
3002
3003 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3004
3005 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3006         struct tgid_iter iter)
3007 {
3008         char name[PROC_NUMBUF];
3009         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3010         return proc_fill_cache(filp, dirent, filldir, name, len,
3011                                 proc_pid_instantiate, iter.task, NULL);
3012 }
3013
3014 /* for the /proc/ directory itself, after non-process stuff has been done */
3015 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3016 {
3017         unsigned int nr;
3018         struct task_struct *reaper;
3019         struct tgid_iter iter;
3020         struct pid_namespace *ns;
3021
3022         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3023                 goto out_no_task;
3024         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3025
3026         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3027         if (!reaper)
3028                 goto out_no_task;
3029
3030         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3031                 const struct pid_entry *p = &proc_base_stuff[nr];
3032                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3033                         goto out;
3034         }
3035
3036         ns = filp->f_dentry->d_sb->s_fs_info;
3037         iter.task = NULL;
3038         iter.tgid = filp->f_pos - TGID_OFFSET;
3039         for (iter = next_tgid(ns, iter);
3040              iter.task;
3041              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3042                 filp->f_pos = iter.tgid + TGID_OFFSET;
3043                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3044                         put_task_struct(iter.task);
3045                         goto out;
3046                 }
3047         }
3048         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3049 out:
3050         put_task_struct(reaper);
3051 out_no_task:
3052         return 0;
3053 }
3054
3055 /*
3056  * Tasks
3057  */
3058 static const struct pid_entry tid_base_stuff[] = {
3059         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3060         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3061         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3062         REG("environ",   S_IRUSR, proc_environ_operations),
3063         INF("auxv",      S_IRUSR, proc_pid_auxv),
3064         ONE("status",    S_IRUGO, proc_pid_status),
3065         ONE("personality", S_IRUGO, proc_pid_personality),
3066         INF("limits",    S_IRUGO, proc_pid_limits),
3067 #ifdef CONFIG_SCHED_DEBUG
3068         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3069 #endif
3070         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3071 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3072         INF("syscall",   S_IRUGO, proc_pid_syscall),
3073 #endif
3074         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3075         ONE("stat",      S_IRUGO, proc_tid_stat),
3076         ONE("statm",     S_IRUGO, proc_pid_statm),
3077         REG("maps",      S_IRUGO, proc_maps_operations),
3078 #ifdef CONFIG_NUMA
3079         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3080 #endif
3081         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3082         LNK("cwd",       proc_cwd_link),
3083         LNK("root",      proc_root_link),
3084         LNK("exe",       proc_exe_link),
3085         REG("mounts",    S_IRUGO, proc_mounts_operations),
3086         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3087 #ifdef CONFIG_PROC_PAGE_MONITOR
3088         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3089         REG("smaps",     S_IRUGO, proc_smaps_operations),
3090         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3091 #endif
3092 #ifdef CONFIG_SECURITY
3093         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3094 #endif
3095 #ifdef CONFIG_KALLSYMS
3096         INF("wchan",     S_IRUGO, proc_pid_wchan),
3097 #endif
3098 #ifdef CONFIG_STACKTRACE
3099         ONE("stack",      S_IRUGO, proc_pid_stack),
3100 #endif
3101 #ifdef CONFIG_SCHEDSTATS
3102         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3103 #endif
3104 #ifdef CONFIG_LATENCYTOP
3105         REG("latency",  S_IRUGO, proc_lstats_operations),
3106 #endif
3107 #ifdef CONFIG_PROC_PID_CPUSET
3108         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3109 #endif
3110 #ifdef CONFIG_CGROUPS
3111         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3112 #endif
3113         INF("oom_score", S_IRUGO, proc_oom_score),
3114         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3115         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3116 #ifdef CONFIG_AUDITSYSCALL
3117         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3118         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3119 #endif
3120 #ifdef CONFIG_FAULT_INJECTION
3121         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3122 #endif
3123 #ifdef CONFIG_TASK_IO_ACCOUNTING
3124         INF("io",       S_IRUSR, proc_tid_io_accounting),
3125 #endif
3126 #ifdef CONFIG_HARDWALL
3127         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3128 #endif
3129 };
3130
3131 static int proc_tid_base_readdir(struct file * filp,
3132                              void * dirent, filldir_t filldir)
3133 {
3134         return proc_pident_readdir(filp,dirent,filldir,
3135                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3136 }
3137
3138 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3139         return proc_pident_lookup(dir, dentry,
3140                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3141 }
3142
3143 static const struct file_operations proc_tid_base_operations = {
3144         .read           = generic_read_dir,
3145         .readdir        = proc_tid_base_readdir,
3146         .llseek         = default_llseek,
3147 };
3148
3149 static const struct inode_operations proc_tid_base_inode_operations = {
3150         .lookup         = proc_tid_base_lookup,
3151         .getattr        = pid_getattr,
3152         .setattr        = proc_setattr,
3153 };
3154
3155 static struct dentry *proc_task_instantiate(struct inode *dir,
3156         struct dentry *dentry, struct task_struct *task, const void *ptr)
3157 {
3158         struct dentry *error = ERR_PTR(-ENOENT);
3159         struct inode *inode;
3160         inode = proc_pid_make_inode(dir->i_sb, task);
3161
3162         if (!inode)
3163                 goto out;
3164         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3165         inode->i_op = &proc_tid_base_inode_operations;
3166         inode->i_fop = &proc_tid_base_operations;
3167         inode->i_flags|=S_IMMUTABLE;
3168
3169         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3170                                                   ARRAY_SIZE(tid_base_stuff)));
3171
3172         d_set_d_op(dentry, &pid_dentry_operations);
3173
3174         d_add(dentry, inode);
3175         /* Close the race of the process dying before we return the dentry */
3176         if (pid_revalidate(dentry, NULL))
3177                 error = NULL;
3178 out:
3179         return error;
3180 }
3181
3182 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3183 {
3184         struct dentry *result = ERR_PTR(-ENOENT);
3185         struct task_struct *task;
3186         struct task_struct *leader = get_proc_task(dir);
3187         unsigned tid;
3188         struct pid_namespace *ns;
3189
3190         if (!leader)
3191                 goto out_no_task;
3192
3193         tid = name_to_int(dentry);
3194         if (tid == ~0U)
3195                 goto out;
3196
3197         ns = dentry->d_sb->s_fs_info;
3198         rcu_read_lock();
3199         task = find_task_by_pid_ns(tid, ns);
3200         if (task)
3201                 get_task_struct(task);
3202         rcu_read_unlock();
3203         if (!task)
3204                 goto out;
3205         if (!same_thread_group(leader, task))
3206                 goto out_drop_task;
3207
3208         result = proc_task_instantiate(dir, dentry, task, NULL);
3209 out_drop_task:
3210         put_task_struct(task);
3211 out:
3212         put_task_struct(leader);
3213 out_no_task:
3214         return result;
3215 }
3216
3217 /*
3218  * Find the first tid of a thread group to return to user space.
3219  *
3220  * Usually this is just the thread group leader, but if the users
3221  * buffer was too small or there was a seek into the middle of the
3222  * directory we have more work todo.
3223  *
3224  * In the case of a short read we start with find_task_by_pid.
3225  *
3226  * In the case of a seek we start with the leader and walk nr
3227  * threads past it.
3228  */
3229 static struct task_struct *first_tid(struct task_struct *leader,
3230                 int tid, int nr, struct pid_namespace *ns)
3231 {
3232         struct task_struct *pos;
3233
3234         rcu_read_lock();
3235         /* Attempt to start with the pid of a thread */
3236         if (tid && (nr > 0)) {
3237                 pos = find_task_by_pid_ns(tid, ns);
3238                 if (pos && (pos->group_leader == leader))
3239                         goto found;
3240         }
3241
3242         /* If nr exceeds the number of threads there is nothing todo */
3243         pos = NULL;
3244         if (nr && nr >= get_nr_threads(leader))
3245                 goto out;
3246
3247         /* If we haven't found our starting place yet start
3248          * with the leader and walk nr threads forward.
3249          */
3250         for (pos = leader; nr > 0; --nr) {
3251                 pos = next_thread(pos);
3252                 if (pos == leader) {
3253                         pos = NULL;
3254                         goto out;
3255                 }
3256         }
3257 found:
3258         get_task_struct(pos);
3259 out:
3260         rcu_read_unlock();
3261         return pos;
3262 }
3263
3264 /*
3265  * Find the next thread in the thread list.
3266  * Return NULL if there is an error or no next thread.
3267  *
3268  * The reference to the input task_struct is released.
3269  */
3270 static struct task_struct *next_tid(struct task_struct *start)
3271 {
3272         struct task_struct *pos = NULL;
3273         rcu_read_lock();
3274         if (pid_alive(start)) {
3275                 pos = next_thread(start);
3276                 if (thread_group_leader(pos))
3277                         pos = NULL;
3278                 else
3279                         get_task_struct(pos);
3280         }
3281         rcu_read_unlock();
3282         put_task_struct(start);
3283         return pos;
3284 }
3285
3286 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3287         struct task_struct *task, int tid)
3288 {
3289         char name[PROC_NUMBUF];
3290         int len = snprintf(name, sizeof(name), "%d", tid);
3291         return proc_fill_cache(filp, dirent, filldir, name, len,
3292                                 proc_task_instantiate, task, NULL);
3293 }
3294
3295 /* for the /proc/TGID/task/ directories */
3296 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3297 {
3298         struct dentry *dentry = filp->f_path.dentry;
3299         struct inode *inode = dentry->d_inode;
3300         struct task_struct *leader = NULL;
3301         struct task_struct *task;
3302         int retval = -ENOENT;
3303         ino_t ino;
3304         int tid;
3305         struct pid_namespace *ns;
3306
3307         task = get_proc_task(inode);
3308         if (!task)
3309                 goto out_no_task;
3310         rcu_read_lock();
3311         if (pid_alive(task)) {
3312                 leader = task->group_leader;
3313                 get_task_struct(leader);
3314         }
3315         rcu_read_unlock();
3316         put_task_struct(task);
3317         if (!leader)
3318                 goto out_no_task;
3319         retval = 0;
3320
3321         switch ((unsigned long)filp->f_pos) {
3322         case 0:
3323                 ino = inode->i_ino;
3324                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3325                         goto out;
3326                 filp->f_pos++;
3327                 /* fall through */
3328         case 1:
3329                 ino = parent_ino(dentry);
3330                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3331                         goto out;
3332                 filp->f_pos++;
3333                 /* fall through */
3334         }
3335
3336         /* f_version caches the tgid value that the last readdir call couldn't
3337          * return. lseek aka telldir automagically resets f_version to 0.
3338          */
3339         ns = filp->f_dentry->d_sb->s_fs_info;
3340         tid = (int)filp->f_version;
3341         filp->f_version = 0;
3342         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3343              task;
3344              task = next_tid(task), filp->f_pos++) {
3345                 tid = task_pid_nr_ns(task, ns);
3346                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3347                         /* returning this tgid failed, save it as the first
3348                          * pid for the next readir call */
3349                         filp->f_version = (u64)tid;
3350                         put_task_struct(task);
3351                         break;
3352                 }
3353         }
3354 out:
3355         put_task_struct(leader);
3356 out_no_task:
3357         return retval;
3358 }
3359
3360 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3361 {
3362         struct inode *inode = dentry->d_inode;
3363         struct task_struct *p = get_proc_task(inode);
3364         generic_fillattr(inode, stat);
3365
3366         if (p) {
3367                 stat->nlink += get_nr_threads(p);
3368                 put_task_struct(p);
3369         }
3370
3371         return 0;
3372 }
3373
3374 static const struct inode_operations proc_task_inode_operations = {
3375         .lookup         = proc_task_lookup,
3376         .getattr        = proc_task_getattr,
3377         .setattr        = proc_setattr,
3378 };
3379
3380 static const struct file_operations proc_task_operations = {
3381         .read           = generic_read_dir,
3382         .readdir        = proc_task_readdir,
3383         .llseek         = default_llseek,
3384 };